LLVM was a thing that demonstrated its value by working when you used it. And you can't judge a population (all non-PhD theses) by its tippety-top performers, particularly when there are poor incentives involved.